In recent years, a computer is used not only in a computer system but also in a mechanical system. For instance, as for the mechanical system such as an automobile, an airplane or production facilities, the computer measures internal states of each component of the mechanical system with various sensors, and automatically controls the components according to measurement results. By way of example, an automobile in recent years has 50 to 100 sensors therein.
As the number of the sensors increases and control becomes complicated. It becomes difficult, in the case of occurrence of trouble to the automobile, to figure out a cause thereof. In particular, in the case when trouble occurs due to a problem of software for the control, or in the case when reoccurence ratio of the trouble is low, there is a limit to repairs made by an engineer at a repair plant depending on the engineer's experience or intuition. For this reason, one measure to figure out the cause of the trouble is by analyzing time-series changes of parameters measured by the sensors. However, it is not easy to properly analyze an enormous amount of data.
The following documents are considered:
[Non-Patent Document 1]                Daxin Jiang, Jian Pei, and Aidong Zhang “DHC: A Density-based Hierarchical Clustering Method for Time Series Gene Expression Data,” Third IEEE Symposium on BioInformatics and BioEngineering (BIBE′ 03)        
[Non-Patent Document 2]                Davood Rafiel, “Fourier-transform based techniques in efficient retrieval of similar time sequences,” Univ. Toronto dissertation, 1999        
[Non-Patent Document 3]                Antonello Panuccio, Manuele Bicego, and Vittorio Murino, “A Hidden Markov Model-based approach to sequential data clustering,” Structural, Syntactic, and Statistical Pattern Recognition, Proceedings of Joint IAPR International Workshops SSPR 2002 and SPR 2002, Windsor, Ontario, Canada, Aug. 6–9, 2002        
[Non-Patent Document 4]                Takehisa Yairi, Yoshikiyo Kato, Koichi Hori, and Shin-ichi Nakasuga, “Method of Malfunction Detection from Artificial Satellite Telemetry Data based on Time Series Correlation Rule Mining,” The 15th Annual Conference on Japanese Society for Artificial Intelligence, 2001, 3D1-01        
[Non-Patent Document 5]                Eamonn Keogh, Jessica Lin, and Wagner Truppel, “Clustering of Time Series Subsequences is Meaningless: Implications for Previous and Future Research,” IEEE International Conference on Data Mining (ICDM 2003)        
[Non-Patent Document 6]                K. Yamaishi and J. Takeuchi, “A Unifying Framework for Detecting Outliers and Change Points from Non-Stationary Time Series Data,” Proc. Of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ACM Press (KDD2002), 2002        
[Non-Patent Document 7]                M. Ghil, M. R. Allen, M. D. Dettinger, K. Ide, D. Kondrashov, M. E. Mann, A. W. Robertson, A Saunders, Y. Tian, F. Varadi, and P. Yiou, “Advanced Spectral Methods for Climatic Time Series,” Reviews of Geophysics, 40 (2002), pp. 1–41        
[Non-Patent Document 8]                N. Golyandina, V. Nekrutkin and A. Zhigljavsky, Analysis of Time Series Structure: SSA and Related Techniques, Chapman and Hall/CRC, 2001        
There has been proposed, for the sake of such an analysis, of a technique for classifying a plurality of measured parameters by groups of the parameters of which behavior is similar to one another (refer to Non-Patent Documents 1 to 4). Non-Patent Documents 5 to 7 will be described later.
In recent years, the computer is not only used in the mechanical system but is also used for calculation and analyses of indices in economy and industries. For instance, a trader dealing in securities such as stocks analyzes economic indices and so on by using the computer to determine timing in buying and selling the securities.
The techniques in Non-Patent Documents 1 to 4 classify parameters based on a tendency of a change in the case where each parameter constantly changes according to passage of time. For instance, time-series data is Fourier-transformed so as to compare cyclical changes of the parameters and classify the parameters showing the same tendency in the same group.
In the case where a first parameter changes suddenly rather than cyclically for instance, these techniques cannot properly detect another parameter influencing the first parameter or still another parameter influenced by the first parameter. For instance, Non-Patent Document 5 points out that the techniques in Non-Patent Documents 1 to 4 are substantially effective only in the case where a parameter constantly changes.
Furthermore, the techniques in Non-Patent Documents 1 to 4 can properly classify the parameters only in the case where the parameters are of mutually similar types. To be more specific, these techniques assume uniformity of parameter values. For this reason, in the case where one parameter is a continuous value and another parameter is a discrete value, it is not possible to determine whether or not the changes are similar as to these parameters. It is not possible to determine whether or not the parameters in mutually different units or ranges are similar. The technique in Non-Patent Document 6 assumes the uniformity of the space direction although it does not assume constancy in a time base direction.
Tsame problem occurs in forecasting price fluctuation of securities, wherein it is difficult to properly determine association between a price of a stock and another index having a measurment unit different from that of the price.